每周一起读 #02 | ICML 2019：基于粒子的变分推断加速方法
”每周一起讀“是由 PaperWeekly 發(fā)起的論文共讀活動(dòng)，我們結(jié)合自然語(yǔ)言處理、計(jì)算機(jī)視覺(jué)和機(jī)器學(xué)習(xí)等領(lǐng)域的頂會(huì)論文和前沿成果來(lái)指定每期論文，并且邀請(qǐng)論文作者來(lái)到現(xiàn)場(chǎng)，和大家展開(kāi)更有價(jià)值的延伸討論。
我們希望能為 PaperWeekly 的各位讀者帶來(lái)一種全新的論文閱讀體驗(yàn)、一個(gè)認(rèn)識(shí)同好、找到組織的契機(jī)、一次與國(guó)際頂會(huì)論文作者當(dāng)面交流的機(jī)會(huì)。
6 月 5 日（周三）晚 7 點(diǎn)半，“每周一起讀”將邀請(qǐng)清華大學(xué)計(jì)算機(jī)系博士生劉暢，和大家分享他發(fā)表于機(jī)器學(xué)習(xí)國(guó)際會(huì)議 ICML 2019 的兩篇最新文章。
? 劉暢??
清華大學(xué)計(jì)算機(jī)系博士生
劉暢，清華大學(xué)計(jì)算機(jī)系博士生，從事統(tǒng)計(jì)機(jī)器學(xué)習(xí)方向研究，導(dǎo)師為朱軍教授。他于 2014 年在清華大學(xué)物理系取得理學(xué)學(xué)士學(xué)位，博士期間曾在杜克大學(xué)訪學(xué)一年。他的研究興趣主要在貝葉斯推理方法以及利用幾何結(jié)構(gòu)的機(jī)器學(xué)習(xí)方法。他在機(jī)器學(xué)習(xí)國(guó)際會(huì)議 ICML, NeurlPS, AAAI 等上發(fā)表了數(shù)篇論文。
? ICML 2019??
Abstract: Particle-based variational inference methods (ParVIs) have gained attention in the Bayesian inference literature, for their capacity to yield flexible and accurate approximations. We explore ParVIs from the perspective of Wasserstein gradient flows, and make both theoretical and practical contributions. We unify various finite-particle approximations that existing ParVIs use, and recognize that the approximation is essentially a compulsory smoothing treatment, in either of two equivalent forms. This novel understanding reveals the assumptions and relations of existing ParVIs, and also inspires new ParVIs. We propose an acceleration framework and a principled bandwidth-selection method for general ParVIs; these are based on the developed theory and leverage the geometry of the Wasserstein space. Experimental results show the improved convergence by the acceleration framework and enhanced sample accuracy by the bandwidth-selection method.
Abstract:?It is known that the Langevin dynamics used in MCMC is the gradient flow of the KL divergence on the Wasserstein space, which helps convergence analysis and inspires recent particle-based variational inference methods (ParVIs). But no more MCMC dynamics is understood in this way. In this work, by developing novel concepts, we propose a theoretical framework that recognizes a general MCMC dynamics as the fiber-gradient Hamiltonian flow on the Wasserstein space of a fiber-Riemannian Poisson manifold. The “conservation + convergence” structure of the flow gives a clear picture on the behavior of general MCMC dynamics. The framework also enables ParVI simulation of MCMC dynamics, which enriches the ParVI family with more efficient dynamics, and also adapts ParVI advantages to MCMCs. We develop two ParVI methods for a particular MCMC dynamics and demonstrate the benefits in experiments.
時(shí)間：6 月 5 日（周三） 19:30–21:00
地點(diǎn)：北京智源人工智能研究院6號(hào)會(huì)議室
北京市海淀區(qū)中關(guān)村南大街1-1號(hào)?
中關(guān)村領(lǐng)創(chuàng)空間（信息谷）
長(zhǎng)按識(shí)別二維碼，即刻報(bào)名?
報(bào)名截止日期：6?月 5 日（周三）12:00
* 場(chǎng)地人數(shù)有限，報(bào)名成功的讀者將收到包含電子門(mén)票二維碼的短信通知，請(qǐng)留意查收。
注意事項(xiàng)：
*?如您無(wú)法按時(shí)到場(chǎng)參與活動(dòng)，請(qǐng)于活動(dòng)開(kāi)始前 24 小時(shí)在 PaperWeekly 微信公眾號(hào)后臺(tái)留言告知，留言格式為放棄報(bào)名 + 報(bào)名電話；無(wú)故缺席者，將不再享有后續(xù)活動(dòng)的報(bào)名資格。
?1 / 掃碼關(guān)注?
掃碼關(guān)注 PaperWeekly?
PaperWeekly
清華大學(xué)計(jì)算機(jī)科學(xué)與技術(shù)系
北京智源人工智能研究院
?
現(xiàn)在，在「知乎」也能找到我們了
進(jìn)入知乎首頁(yè)搜索「PaperWeekly」
點(diǎn)擊「關(guān)注」訂閱我們的專(zhuān)欄吧
關(guān)于PaperWeekly
PaperWeekly 是一個(gè)推薦、解讀、討論、報(bào)道人工智能前沿論文成果的學(xué)術(shù)平臺(tái)。如果你研究或從事 AI 領(lǐng)域，歡迎在公眾號(hào)后臺(tái)點(diǎn)擊「交流群」，小助手將把你帶入 PaperWeekly 的交流群里。
▽ 點(diǎn)擊 |?閱讀原文?| 立刻報(bào)名
與50位技術(shù)專(zhuān)家面對(duì)面20年技術(shù)見(jiàn)證，附贈(zèng)技術(shù)全景圖總結(jié)
以上是生活随笔為你收集整理的每周一起读 #02 | ICML 2019：基于粒子的变分推断加速方法的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。
- 上一篇： 震撼来袭 | 人工智能Paper精读班，
- 下一篇： ACL 2019 | 基于知识增强的语言